The present disclosure relates to a medical image processing apparatus, method, and program for recognizing a vertebra included in a medical image.
A spinal cord is a very important region as it plays a role of conveying messages between a brain and each body region. For this reason, the spinal cord is protected by a plurality of vertebrae (a spine). In addition, it is practiced to confirm the presence or absence of damage or a lesion in a vertebra by interpreting a tomographic image obtained by scanning a subject. At this time, it is necessary to identify each vertebra, for example, to report a damaged or lesioned vertebra. Therefore, various image processing algorithms for separating and recognizing a plurality of vertebrae based on a tomographic image obtained by scanning a subject and labeling each vertebra.
For example, Japanese Unexamined Patent Publication No. 2011-131040 proposes a method for generating, with a three-dimensional image obtained from tomographic images of a computed tomography (CT) image, a magnetic resonance imaging (MRI) image, or the like as the target, tomographic images of planes intersecting and parallel to the central axis of each vertebra, calculating a characteristic amount representing sharpness of a cross-sectional shape in each tomographic image and a characteristic amount representing regularity of an array of vertebrae, identifying an area of each vertebra by identifying a position of an intervertebral disc located between each vertebra based on these characteristic amounts, and further labeling the identified area of each vertebra.
The literature M. Lootus et al., “Vertebrae Detection and Labelling in Lumbar MR Images”, MICCAI Workshop: Computational Methods and Clinical Applications for Spine Imaging, 2013 proposes a method of extrapolating vertebrae and a sacrum using Deformable Part Model and Dynamic Programming. The method described in the aforementioned literature allows extrapolation of a sacrum and labeling of vertebrae with reference to the sacrum.